Branch and Vault
Live
Payments

Agentic AI Reshapes Institutional Payments with Bounded Execution

Deutsche Bank and Ant Digital Technologies outline a governance-led framework for AI agents in wholesale banking, focusing on treasury, payments, and trade finance under a three-layer IMF architecture.

Deutsche Bank and Ant Digital Technologies outline a governance-led framework for AI agents in wholesale banking...

Deutsche Bank and Ant Digital Technologies released a white paper titled 'Agentic Payments in Wholesale Finance' on November 9, 2026. The research explores how agentic AI can automate institutional treasury, cross-border payments, and trade finance under strict governance controls. It draws on a three-layer reference architecture proposed by the International Monetary Fund (IMF) to structure how agents reason, are constrained, and execute transactions.

The three-layer IMF reference architecture structures how agents reason, are constrained, and execute transactions. Intent and Orchestration oversees adaptive agent reasoning, decision-making and workflow orchestration. Control and Authorisation enforces deterministic, rule-based constraints and ensures that actions remain within predefined authorisation boundaries. Settlement carries out the final execution of financial transactions.

What agents can do in treasury and trade

AI agents can autonomously forecast liquidity, orchestrate cross-border payments, and streamline trade finance under bounded mandates. The report explores how these agents can autonomously forecast liquidity shortfalls and surpluses for a treasury department. They assess available response options and execute compliant actions across internal fund transfers, investments, external financing and FX operations.

In cross-border payment orchestration, an agent could weigh FX timing, payment rails, fees and settlement risk against client objectives. For trade finance, AI agents can interpret complex trade documentation, identify data discrepancies, coordinate exception handling and generate financing or payment instructions once conditions are satisfied. They could convert unstructured documents into traceable evidence and run sanctions, AML and credit checks.

Governance and the human-in-the-loop

Authorisation limits, Know Your Agent protocols, and deterministic controls ensure agents propose but do not execute beyond approved boundaries. The white paper advocates an approach based on 'bounded agentic execution' rather than unrestricted autonomy. The separation between AI-driven reasoning and the execution of financial transactions allows for essential safeguards. These include authorisation limits, agent identity binding, and comprehensive Know Your Agent (KYA) verification protocols.

An agent may propose a payment, but an independent control layer owned by the institution decides whether it proceeds. The design principle is that the agent supplies judgement while the institution supplies permission. Signing keys sit outside the agent's reasoning environment, mandates are machine-readable and revocable in near real time. High-value or exceptional transactions still route to a named human. The paper ties this to EU AI Act Article 26(2), requiring deployers of high-risk AI systems to assign oversight to competent, trained individuals.

Sabih Behzad, head of digital assets and currencies transformation at Deutsche Bank, stated the potential value lies not simply in executing payments automatically, but in improving how those decisions are made, coordinated, and controlled.

Real-world illustration and market context

Anvita Flow demonstrates the framework, while industry adoption is accelerating alongside broader agent market growth. Anvita Flow, an advanced agentic platform developed by Ant Digital Technologies, is featured as a key implementation case study. It illustrates how financial institutions can orchestrate complex payment workflows securely.

The platform binds each agent to a verified owner through know-your-agent checks and caps spending and counterparties in a scoped mandate. It grades settlement assurance to value at risk, from roughly 100 milliseconds for low-value activity to minute-scale proof for material transfers. Zhuoqun Bian, president of Ant Digital Technologies, stated that realising this potential in wholesale finance depends on foundations that are neither optional nor purely technical. Agents need verifiable identity, bound to an accountable owner. Their authority must be explicit, machine-readable, bounded, and revocable.

The World Economic Forum has estimated the agent market could be worth 236 billion US dollars by 2034. The white paper's central recommendation is for institutions to remain in the middle of the automation spectrum, near AI-assisted decision-making with bounded agentic execution.

Topics

#Payments

Related coverage

More from Payments